Works with the AI Agents you already use
Stale Claim Guard: Catch Dead Facts in Your AI Agent's Memory Files
Checks the [verify | checked: date] tags in CLAUDE.md, AGENTS.md, notes and skills (too old, malformed, future-dated), lists every file read N of N and exits 2, so stale memory cannot reach your agent's answers.
$3
Stale Claim Guard: Catch Dead Facts in Your AI Agent's Memory Files
Example session with this skill installed
Before we answer anything from our notes, check CLAUDE.md and the notes/ folder for stale facts.
- Read your context and instructions
- Compiled the stale claim guard
FINDINGS notes/infra.md
line 3 [old-tag] [verify: ls /backup | checked: 2026-09-01] (67 days old, limit 14)
line 4 [bad-tag] [verify: nothing]
files examined 12 of 12 (ok 11, with findings 1, could not read 0)
findings: 2
RESULT: not clean. (exit 2)
Connects securely to your tools. The creator never sees your data.
About this skill
A memory file is a hypothesis with a timestamp. Rules last. Decisions last for a while. State does not: "we have no staging server", "the invoice is not yet sent", "awaiting the client's signature", "the build is deployed". Each was true on the day it was written. An agent that reads it three weeks later repeats it with full confidence, and the person on the other side acts on it.
Stale Claim Guard gives every state claim a shape that can be re-checked, and checks that shape for you:
[verify: <how to re-check it> | checked: YYYY-MM-DD]
What it does by default (exact):
- Walks the files and folders you give it (CLAUDE.md, AGENTS.md, notes, skills), following symlinked folders once.
- Flags every verification tag that is older than your limit (14 days by default), malformed, or dated in the future.
- Prints one state per file (OK, FINDINGS, COULD NOT READ) and always "files examined N of N". An unreadable file is never skipped silently.
- Exit 0 clean, exit 2 findings or anything unread: ready for a session-start check, a pre-commit hook or CI.
Optional, experimental: --claims also lists sentences that read like current-state claims with no tag at all. It matches words, so treat it as a reading list, not a verdict.
One Python 3 script, standard library only, no network, writes nothing. Ships with a 53-case selftest you run first.
Built from a real pattern: in one month, five confident wrong answers came out of honest notes that had outlived their truth (an account "that did not exist", a document "awaiting signature" already sent, a form "never submitted" that had gone out the night before).
It does not check whether a claim is true, only that it says how to be re-checked and when it last was. The --claims mode is experimental: it matches words, so it also lists rules and lessons; treat it as a reading list. English only.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 9 days ago
Requires
Permissions
File Scopes
No network calls, reads no secrets, writes nothing.